<html> <head> <base href="https://bugs.freedesktop.org/"> </head> <body> <div> <a class="bz_bug_link bz_status_NEW " title="NEW - [bdw-u iommu] DMAR error -> GPU hang" href="https://bugs.freedesktop.org/show_bug.cgi?id=89360#c58">Comment # 58</a> on <a class="bz_bug_link bz_status_NEW " title="NEW - [bdw-u iommu] DMAR error -> GPU hang" href="https://bugs.freedesktop.org/show_bug.cgi?id=89360">bug 89360</a> from <a class="email" href="mailto:changbin.du@intel.com" title="Du, Changbin <changbin.du@intel.com>"> Du, Changbin</a> <pre>Chris, I did a experiment that make the intel_unmap as a noop. Then the old dma region keeps mapped, and new allocated buf always has new IOVA. Thus I didn't reproduce this issue. Therefor, I guess BDW GPU may still issue DMA transaction with old unmapped dma region, even that workload has finished. And I always see the hang at PIPE_CTL. --- a/drivers/iommu/intel-iommu.c +++ b/drivers/iommu/intel-iommu.c @@ -3787,7 +3787,7 @@ static void intel_unmap(struct device *dev, dma_addr_t dev_addr, size_t size) struct intel_iommu *iommu; struct page *freelist; - if (iommu_no_mapping(dev)) + //if (iommu_no_mapping(dev)) return; domain = find_domain(dev); BTW, sometimes the IOMMU fault addr is 48 bit which more like a Gfx VA, but sometimes it is 33bit or less which more like a IOVA or Pysical Address. Per my understanding IOMMU fault should always be IOVA, so how does a Gfx VA recored, is the dedicated gfx iommu special? Thanks, Changbin.</pre> </div> <hr> You are receiving this mail because: <ul> <li>You are the QA Contact for the bug.</li> <li>You are the assignee for the bug.</li> <li>You are on the CC list for the bug.</li> </ul> </body> </html>